Short-term forecast of high-energy electron flux based on GPR

The rapid enhancement of high-energy electron flux affects the safe operation of satellites in the synchronous orbit area, so accurate forecast of flux is a key focus in space weather research. This study uses >2 MeV electron flux, solar wind parameters, and geomagnetic parameters from 2001 to 2006 to perform a delayed analysis of input parameters and establish a prediction model named gaussian process regression (GPR) based on machine learning, where sets of 2001-2005 are used as training and sets from January to December in 2006 as testing. It is shown that the GPR is found to have a better performance when comparing with four typical and widely used models: RDF, Low-E, FLUXPRED, and REFM. It also outperforms other intelligence models like backward propagation neural network (BPNN), support vector machine regression (SVR), decision tree regression (DT), and long short-term memory network (LSTM) in terms of flux forecast. The prediction efficiency (PE), correlation coefficient (R), and root mean square error (RMSE) of our GPR model are 0.83, 0.91, and 0.39, respectively. The validation of GPR is further verified by its relatively better prediction of extreme disturbed events in which electron flux suddenly increased or decreased by several orders of magnitude.